We propose efficient and effective functionally equivalent model extraction attacks against tree-based models, along with new formal analysis methods to characterize their efficiency.